Strain engineering (synbio CRO)

01Overview and value chain

Markers: [EC: EU industrial-biotechnology programmes / TRL 1–6 scale-up | OECD: 2.8 Industrial biotechnology | Regulator: EPA (US)]

A strain-engineering contract research organisation sells the biological catalyst rather than the product: given a target reaction or molecule, it delivers an enzyme variant or a production strain that makes the process economic. The work is organised as a design-build-test-learn cycle, and the modern differentiator is what trains the design step — proprietary experimental datasets accumulated over fifteen to twenty years rather than public sequence databases alone. The gains are measurable at process level: in one documented pharmaceutical cascade the engineered enzymes raised substrate loading from 5 g/L to 67 g/L, cut the enzyme load of one enzyme from 100 weight percent to 8 and of another from 100 to 3, shortened reaction time from 21 hours to 12, and held stereochemical purity above 99% enantiomeric and diastereomeric excess. That is the value proposition in a sentence: the same chemistry, run at an order of magnitude higher concentration with a fraction of the catalyst. The sector splits into commercial platform companies selling engineered enzymes, public-private biofoundries carrying projects from research to pre-industrial scale, and large gene-synthesis houses that grew from oligonucleotides into full molecular-biology service stacks.

The key directions of strain engineering services are:

1. Directed Evolution Platforms: integrated enzyme-engineering platforms combining machine learning, bioinformatics and laboratory workflows in a closed design-to-validation loop, trained on two decades of in-house experimental data rather than public data alone.
2. Computational Protein Design: physics-based de novo design driven by AI agents over a proprietary structured dataset, aiming at hit rates ten to a hundred times higher than mutagenesis-led screening, and carried through to manufacturability and formulation rather than stopping at a sequence.
3. Genome Editing of Production Strains: CRISPR nuclease families developed specifically to edit industrial hosts — E. coli, Bacillus, Pichia, Aspergillus — with activity across prokaryotic and eukaryotic organisms, used to optimise the chassis rather than to treat disease.
4. Public-Private Biofoundries: consortium-funded facilities that carry microbial cell-factory projects across TRL 1 to 6, covering synthetic biology, microbial resource management, bioprocess development and scale-up in one place.

Sectoral value chain

[Target Definition] ──> [Design] ──> [Build & Screen] ──> [Strain/Enzyme Delivery]
                                              │
                                      (test data feedback)
                                              │
                                              ▼
[Industrial Process] <─── [Bioprocess Scale-Up] <─────────┘

Value chain levels

Cross-cutting technologies of the sector:

• Proprietary Design Datasets: the structured record of what has actually been built and measured in-house over fifteen to twenty years, which is what separates a design platform from a public-database screen.
• Closed-Loop Design-Build-Test-Learn: the engineering cycle that turns each screening round into training data for the next design round, increasingly driven by AI agents rather than manual iteration.
• Industrial-Host Genome Editing: nuclease families selected for activity in production organisms rather than in mammalian cell lines, enabling targeted edits in the chassis that actually makes the product.

02US

The United States holds the commercial platform layer: privately held companies that sell engineered proteins as a product, backed by long-accumulated proprietary datasets and validated through named pharmaceutical partnerships.

Codexis CodeEvolver, Arzeda computational design, documented process gains

• Codexis (CodeEvolver): an integrated, proprietary enzyme-engineering platform combining machine learning, bioinformatics and laboratory workflows in a fully connected design-to-validation workflow, with models trained on more than twenty years of Codexis-specific experimental data rather than public data alone, applied across pharmaceuticals, RNA therapeutics and diagnostics.
• Documented process gains: in a collaboration with Bristol Myers Squibb on BMS-986278, an LPA1 antagonist for pulmonary fibrosis, Codexis engineered a three-enzyme biocatalytic cascade, evolving an ERED and a KRED to hold above 99% enantiomeric and diastereomeric excess while raising substrate loading from 5 g/L to 67 g/L, cutting the ERED load from 100 weight percent to 8 and the KRED from 100 to 3, and shortening reaction time from 21 hours to 12.
• Arzeda (Intelligent Protein Design): AI agents drive a closed-loop design-build-test-learn cycle over a proprietary multi-year dataset with physics-based protein design rather than mutagenesis or public sequences; the company reports more than fifteen years of structured design data and targets hit rates ten to a hundred times higher, carrying designs through productisation, formulation and manufacturability rather than delivering a sequence.

03CN

China’s strain-engineering service base grew out of gene synthesis rather than enzyme evolution: the anchor company scaled from oligonucleotide and gene-synthesis work into a full molecular-biology and biologics service stack, inside a policy environment that now names bio-manufacturing as a priority future industry.

GenScript gene-synthesis origin, full service stack, bio-manufacturing policy priority

• GenScript (金斯瑞): founded in 2002 in New Jersey and built on gene synthesis, it expanded from gene-cloning efficiency, cost and quality work into antibodies, reagents, peptide synthesis, protein expression, antibody and protein engineering, and further into immunotherapy and biologics CDMO operations, serving scientists in more than one hundred countries.
• Domestic synbio CRO sector (qualitative): Chinese industry analysis describes the field organised around the design-build-test-learn cycle applied to chassis cells such as E. coli and Saccharomyces cerevisiae, with the value chain split into upstream read-write-edit technologies, midstream platform companies and downstream applications — but no second domestic strain-engineering CRO was independently confirmed in 2026 sourcing, so the sector is described here qualitatively rather than tabled.
• Policy pull: bio-manufacturing was named first among the future industries in the 2025 Chinese government work report’s commitment to build an investment-growth mechanism for future industries, and the field sustains its own industrial conference series, with the fourth China synthetic-biology and bio-manufacturing conference held in Shenzhen in January 2026.

04EU

Europe’s distinctive contribution is institutional rather than corporate: a public-private biofoundry model that carries projects from research to pre-industrial scale, alongside a listed industrial-biotech company whose genome-editing IP targets production strains directly.

TWB public-private biofoundry, BRAIN Biotech BMC nucleases, TRL 1–6 coverage

• Toulouse White Biotechnology (TWB): a public-private biofoundry in France jointly managed by INRAE, INSA and CNRS, created in 2012 and operating as a joint service unit with a consortium of roughly 45 to 50 members. It covers synthetic biology, microbial resource management, bioprocess development and scale-up across TRL 1 to 6, running around 50 R&D projects a year and having completed more than 400 since inception.
• BRAIN Biotech (BMC nuclease family): its BRAINBiocatalysts segment supplies enzymes, microorganisms and ingredients for industrial use alongside CRO activity and production-strain development, and it holds a granted European patent (EP4301852 B1) for the CRISPR-BMC nuclease. The nuclease is active across prokaryotic and eukaryotic organisms — bacteria, yeasts, fungi, plants and mammalian cells — and is aimed explicitly at optimising microbial production strains such as E. coli, Bacillus, Pichia and Aspergillus.
• The pre-industrial gap: the European model is built around the step that neither a university nor a commercial client wants to own — carrying a validated strain from laboratory result to a bioprocess that can be handed to industry — which is why the biofoundry sits as a shared consortium asset rather than a private service company.

05Leading companies and research institutes

06Tech stack and innovations

The stack is organised around one loop — design, build, test, learn — with the competitive difference lying in what trains the design step and how far the provider carries the result toward manufacture.

1. Directed Evolution with Machine Learning:
   • An integrated platform connects computational design to experimental validation in one workflow, with proprietary models trained on twenty-plus years of the provider’s own experimental results rather than only on public sequence data.
   • The output is measured in process terms: a three-enzyme cascade delivering above 99% enantiomeric and diastereomeric excess while substrate loading rises from 5 g/L to 67 g/L and enzyme loads fall from 100 weight percent to 8 and to 3, with reaction time down from 21 hours to 12.
2. Physics-Based Computational Protein Design:
   • De novo design guided by physics and a proprietary structured dataset, with AI agents running the design-build-test-learn cycle, is positioned against mutagenesis and public-sequence approaches and targets hit rates ten to a hundred times higher.
   • The scope extends past the sequence: designs are validated under application-relevant conditions and taken through formulation and manufacturability, because a protein that cannot be produced or formulated is not a product.
3. Genome Editing Aimed at Industrial Hosts:
   • The BMC nuclease family, protected by a granted European patent, provides high activity across prokaryotic and eukaryotic organisms and is aimed at defined edits in the production chassis — E. coli, Bacillus, Pichia, Aspergillus — rather than at therapeutic editing.
   • This matters because the limiting factor in an industrial fermentation is usually the host’s own metabolism, not the pathway enzyme in isolation.
4. Biofoundry Infrastructure and Scale-Up:
   • A public-private biofoundry spans synthetic biology, microbial resource management, bioprocess development and scale-up in one unit across TRL 1 to 6, sustaining roughly 50 projects a year and more than 400 completed.
   • The consortium structure exists because the pre-industrial step is expensive, shared across roughly 45 to 50 members, and is the stage at which most laboratory results fail to become processes.

07Value chains and production pipelines

Industrial pipeline of a strain-engineering engagement (TRL 1–6 progression, EPA oversight of engineered organisms)

┌───────────────────────────┐      ┌───────────────────────────┐
│ 1. Target Specification   │ ───> │ 2. Computational Design   │
└───────────────────────────┘      └───────────────────────────┘
                                                  │
                                                  ▼
┌───────────────────────────┐      ┌───────────────────────────┐
│ 4. Screening Under        │ <─── │ 3. Build & Expression     │
│    Process Conditions     │      │                           │
└───────────────────────────┘      └───────────────────────────┘
               │
               ▼
┌───────────────────────────┐      ┌───────────────────────────┐
│ 5. Bioprocess Scale-Up    │ ───> │ 6. Handover & Production  │
└───────────────────────────┘      └───────────────────────────┘

Stage 1: Target Specification

The client’s reaction or molecule is translated into a protein or strain specification that includes the conditions the catalyst must survive — substrate concentration, temperature, solvent, cycle time. Specifying against real process conditions rather than assay conditions is what determines whether the eventual variant survives contact with the plant.

Stage 2: Computational Design

Variants are proposed by physics-based de novo design or by machine-learning models trained on the provider’s proprietary dataset — fifteen to twenty years of structured in-house results — rather than by random mutagenesis. This is where the ten- to hundred-fold hit-rate claims are made, and where public-data-only approaches are explicitly said to fall short.

Stage 3: Build and Expression

Designed variants are synthesised and expressed in the chosen chassis. Where the host itself limits performance, genome editing with nucleases selected for activity in E. coli, Bacillus, Pichia or Aspergillus alters the production strain rather than only the pathway enzyme.

Stage 4: Screening Under Process Conditions

Variants are measured under application-relevant conditions, and the results are fed back as training data for the next design round — the closed loop that distinguishes a platform from a one-off screening campaign. The metrics that matter are the industrial ones: substrate loading, catalyst load, stereochemical purity and reaction time.

Stage 5: Bioprocess Scale-Up

The validated biology is carried from bench toward pre-industrial scale, the TRL 1 to 6 span a biofoundry is built to cover. This is the stage the consortium model exists to fund, because it is expensive, shared across roughly 45 to 50 members, and the point at which most laboratory results fail to become processes.

Stage 6: Handover and Production

The engineered enzyme or strain is delivered with its validation data, and the gains appear in the client’s plant as higher substrate loading, lower enzyme load and shorter cycle times — in the documented case, an order-of-magnitude concentration increase with catalyst use cut to a few weight percent, which is what makes a biocatalytic route competitive with a chemical one.